Cognitive Trajectory Changes Over 20 Years Before Dementia Diagnosis: A Large Cohort Study

Abstract

Background/objectives: Longitudinal studies have shown an increase in cognitive decline many years before clinical diagnosis of dementia. We sought to estimate changes, relative to "normal" aging, in the trajectory of scores on a global cognitive function test-the Cognitive Abilities Screening Instrument (CASI).

Design: A prospective cohort study.

Setting: Community-dwelling members of a U.S. health maintenance organization.

Participants: Individuals aged 65 and older who had no dementia diagnosis at baseline and had at least two visits with valid CASI test score (N = 4,315).

Measurements: Average longitudinal trajectories, including changes in trajectory before clinical diagnosis in those who would be diagnosed with dementia, were estimated for CASI item response theory (IRT) scores. The impact of sex, education level, and APOE genotype on cognitive trajectories was assessed.

Results: Increased cognitive decline relative to "normal" aging was evident in CASI IRT at least 10 years before clinical diagnosis. Male gender, lower education, and presence of ≥1 APOE ε4 alleles were associated with lower average IRT scores. In those who would be diagnosed with dementia, a trajectory change point was estimated at an average of 3.1 years (95% confidence interval 3.0-3.2) before clinical diagnosis, after which cognitive decline appeared to accelerate. The change point did not differ by sex, education level, or APOE ε4 genotype. There were subtle differences in trajectory slopes by sex and APOE ε4 genotype, but not by education.

Conclusion: Decline in average global cognitive function was evident at least 10 years before clinical diagnosis of dementia. The decline accelerated about 3 years before clinical diagnosis.

Keywords: change point; cognitive function; dementia; trajectory changes.

Figure 1. CASI IRT trajectories by dementia status

Upper panels (A, B): CASI IRT scores by time to diagnosis for those diagnosed with dementia (A) or time to censoring^† for normal aging (B) for the entire study sample. Solid lines correspond to non-parametric loess smoothed trajectories ^†Records for those not diagnosed with dementia were censored at the next to last visit. Lower panels (C,D): Individual observed and estimated mean age trajectories for those diagnosed with dementia at age 85 (C) or normal aging censored at age 85 (D). For college educated females in the 75 and under baseline age cohort. Solid black line: fitted (piece-wise) linear primary models Dashed black line: fitted fractional polynomial (FP) models The components of the fitted FP model for normal aging are of the form ((age + 10.1)/10)² and ((age + 10.1)/10)³, where age was centered at 75 years, and for time to dementia diagnosis (t) are of the form: ((t + 0.1)/10)^-0.5 and ((t + 0.1)/10)^-0.5 * log(((t + 0.1)/10)).